ScanScript — order-intake proofer for Indian diagnostic labs

1. One-liner

Reads a doctor’s prescription — handwritten or spoken — into an Indian lab’s exact coded test order, before the draw.

2. Trend signal — why now?

Three things landed in the same 12 months, and they point at the same soft spot.

First, the money. India’s diagnostic-lab market is huge and growing double-digit — depending on whose report you believe, ₹90K crore–plus and compounding at ~11% a year. Standalone and independent labs hold ~41–42% of it. At-home collection went from novelty to default: Redcliffe alone runs 2,000+ collection points across 220+ cities, Orange Health promises a phlebotomist at your door in 60 minutes. Every one of those home visits starts with a human reading a doctor’s chit and typing test names into an app.

Second, the pain is now measured and named. Industry write-ups peg the loss at ₹3 lakh/year per lab from “misplaced samples, wrong spellings, and data entry typos,” and put it bluntly: “A missing digit in the MRN or a wrong test code means samples sit idle or are tested incorrectly” (pathlims.com). LIS vendors sell barcode accessioning and digital requisitions specifically to kill “the silent money drains” — but every one of those tools starts after the order is already keyed in correctly. Nobody proofs the keystroke itself.

Third, the feasibility flipped. In 2026 Sarvam shipped Saaras v3 — telephony-tuned speech-to-text for 22 Indian languages at ₹30/hour, built for 8kHz phone audio, with a startup credit program. Reading a mixed Hindi-English voice note or a phone call and pulling out “LFT, TFT, HbA1c, Vitamin D” is now a cheap API call, not a research project. Same for vision: reading a scrawled prescription is squarely inside current multimodal models.

Provenance:

• Signal 1 (Demand): Indian labs lose ~₹3L/yr to manual order-entry errors; “wrong test code means samples sit idle or are tested incorrectly” — https://pathlims.com/blogs/common-lab-errors-and-fixes/ — 2026-07-02
• Signal 2 (Feasibility): Sarvam Saaras v3 telephony STT, 22 Indian languages, ₹30/hr, tuned for 8kHz phone audio + startup credit program — https://www.sarvam.ai/api-pricing — 2026-07-02
• Signal 3 (Economic): India diagnostic-labs market growing ~11% CAGR; standalone/independent labs ~41–42% share; at-home collection scaled (Redcliffe 2,000+ collection points, 220+ cities) — https://www.imarcgroup.com/india-diagnostic-labs-market , https://www.orangehealth.in/ — 2026-07-02 Category: Tech-unlock

3. The opportunity

The gap is a single step everyone skips: turning a free-form doctor’s order into this lab’s exact coded test panel.

Every lab codes its own test master differently. “Thyroid profile” at Lab A is “TFT — T3, T4, TSH”; at Lab B it’s a bundle that also includes anti-TPO. A doctor writes “TFT” in three-second handwriting, or a patient says it over the phone in Hinglish, and a ₹12,000/month front-desk clerk maps it to a code from memory. Get it wrong and you draw the wrong tube, the sample is useless, and you’re calling the patient back for a re-draw — refund, apology, and a one-star review that says “they made me give blood twice.”

The incumbents (Drlogy — 30,000+ labs, Flabs, CrelioHealth, LabSmart) are LIS/billing systems. They give the phlebotomist a requisition screen — but the screen assumes you already know the right code to type. They automate everything downstream of correct intake: accessioning, analyzer interfacing, report delivery on WhatsApp. The one thing they leave to a tired human is the mapping from “what the doctor scrawled” to “what to actually draw and run.” That’s the 10× step an AI-first tool owns: read the script (image or voice), match it against this lab’s master, flag ambiguity, and hand back a clean, confirmed order.

4. Target market

• Primary customer: Owner-operators of standalone diagnostic labs and collection centers in Tier-1/Tier-2 India — 1–10 staff, ₹5–40L/month revenue, running a LIS like Drlogy but still keying orders by hand. Secondary: home-collection aggregators’ phlebotomist ops teams.
• Why they buy: “Every re-draw costs me the reagent, the phlebotomist’s second trip, a refund, and a Google review. My front desk mis-books tests every week and I only find out when the sample’s already spoiled.” The pain is felt in rupees and reputation, weekly.
• Rough TAM reasoning: India has tens of thousands of standalone labs and well over 100,000 collection centers (2,000+ from Redcliffe alone). Even a conservative serviceable base of 15,000–25,000 labs willing to pay a software line-item is a real market for a bootstrapper.
• Why now for them: Home collection exploded their order volume and moved intake off the counter and onto phones and WhatsApp voice notes — exactly where handwriting and accent errors multiply. And the tool to read those is finally cheap in Indian languages.

5. Product sketch (MVP)

• Snap-a-script: phlebotomist or front desk photographs the doctor’s prescription; ScanScript returns the mapped list of tests against this lab’s master, with a confidence flag on each.
• Voice intake: patient dictates tests over phone or WhatsApp voice note (Hindi/English/regional mix); ScanScript transcribes and maps.
• Ambiguity catcher: when “sugar” could mean fasting glucose, PP, or HbA1c, it surfaces the choices instead of guessing — the one moment a human must confirm.
• Lab-specific master mapping: learns each lab’s panel definitions and local shorthand, so “TFT” resolves to their code, not a generic one.
• Duplicate & missing-test check: warns if the same analyte is ordered twice across panels, or if a written test wasn’t captured.
• One-tap confirm → push to LIS: confirmed order exports to Drlogy/Flabs/Excel/WhatsApp so it drops into the existing workflow.
• Error ledger: a running log of caught mis-maps and re-draws avoided — the ROI screen the owner checks.

6. AI angle — what’s load-bearing

Remove the AI and there is no product — it’s just a form the clerk already fills in wrong. The AI does two hard jobs a rules engine can’t: (1) read genuinely messy inputs — doctor handwriting, abbreviations, and accented multilingual voice — and (2) map loose free-text to a specific lab’s idiosyncratic test master while knowing when it’s unsure. That last part matters: a naive model that confidently guesses is worse than a clerk. The value is calibrated confidence — auto-pass the 90% it’s sure about, escalate the 10% it isn’t. Vision for scripts, Sarvam-class telephony STT for voice, and an LLM for the fuzzy master-matching. All load-bearing.

7. Localization angle

This is India-first by construction, not by choice. Doctor prescriptions here are handwritten in English shorthand mixed with local terms; patients dictate in Hinglish, Tamil-English, Marathi-English. Global STT/vision stacks choke on exactly this; Sarvam’s 22-language, telephony-tuned, ₹-priced models are the wedge. Pricing has to be ₹-native — a ₹999–2,999/month tool works where a $49 one is dead on arrival. Distribution is WhatsApp-first because that’s where the orders already arrive. A generic global “prescription reader” would be worse and more expensive here.

8. Business model — path to $1M–$5M ARR

• Pricing: ₹1,499/month base per lab (up to N orders), ₹2,999/month for high-volume labs and small collection-center chains. Usage overage on order volume for aggregators.
• ACV: ₹24,000 ($290) blended.
• Rough math to $1M ARR: ~290 labs × ₹2,499/mo avg × 12 ≈ ₹87L… so realistically ~350–400 paying labs at blended ₹2,000–2,500/mo gets you to ~$1M. Against a base of 15,000+ candidate labs, that’s ~2.5% penetration.
• Rough math to $5M ARR: ~1,700–2,000 labs, or 300–400 labs plus a couple of home-collection aggregator contracts priced per-order (their volume dwarfs a single lab). Aggregators are the ACV multiplier.
• Expansion path: per-order usage as home-collection volume grows; add-on modules — sample-rejection prediction, report-abnormality pre-flagging, patient re-order nudges over WhatsApp. Each rides the same intake data you already hold.

9. Go-to-market wedge — first 100 customers

• Ride the LIS directories. Drlogy, Flabs, LabSmart, Techjockey and SaaSworthy list thousands of labs already shopping for lab software. Scrape the lab-software review pages and Justdial/IndiaMART lab listings in 5 metros; you have a named list of 2,000+ operators in a week.
• Personalized proof, not a pitch. Cold-WhatsApp each lab a 40-second video: photograph a real messy prescription, show ScanScript mapping it to their likely master, flag one ambiguity. “This is the re-draw you had last week.” Expect 3–5% to reply for a free 30-day pilot.
• Phlebotomist WhatsApp groups. Home-collection phlebotomists organize in city-level WhatsApp/Telegram groups. Seed a free personal tier for phlebs (snap-a-script for their own draws); they pull it into the labs they serve.
• Lab-owner associations & equipment dealers. Regional pathology-lab associations and the reagent/analyzer dealers who visit every lab monthly are a warm channel — dealer gets a referral cut.
• One aggregator lighthouse. Land a single mid-size home-collection player as a per-order pilot; their phleb ops team is 100+ intake points in one logo.

10. Build complexity — justification

Medium. The AI is off-the-shelf (Sarvam telephony STT, a multimodal LLM for vision + master-matching) — no model training in v1. The real work is the per-lab test-master onboarding (ingesting each lab’s panel definitions and shorthand) and the confidence-calibration/escalation logic that decides auto-pass vs. ask-a-human. Plus lightweight LIS export connectors (Drlogy/Flabs/Excel/WhatsApp). A 2–3 person team ships a credible v1 in ~3–4 months; the onboarding tooling is what stretches it past a 6-week hack.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Software tool for licensed labs; no diagnosis, human confirms every order.
Ethical — no harm / dark patterns	✅	Reduces mis-draws; keeps human-in-the-loop on ambiguity by design.
Market exists (evidence above)	✅	₹3L/yr loss documented, 15K+ standalone labs, at-home boom.
1–5 person team can build this	✅	Off-the-shelf AI; work is onboarding + connectors.
Launchable with <$50K / ₹40L	✅	No hardware, ₹-priced inference, direct WhatsApp GTM.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	15/20	Real, weekly, costs rupees + reviews — but felt as scattered small losses, not one hair-on-fire bill, and clerks have muddled through for years.
Demand evidence	15	11/15	Sourced ₹3L/yr loss, documented pre-analytical error studies, LIS vendors monetizing adjacent pain. No direct “shut up and take my money” quote for this exact slice yet.
Build feasibility	15	11/15	AI is off-the-shelf; per-lab master onboarding and calibration are the honest 3–4 month cost.
Distribution clarity	15	11/15	Named lists (LIS directories, Justdial, phleb groups) and a concrete proof-video motion; conversion still unproven.
Revenue mechanics	15	11/15	₹-native pricing benchmarked to existing LIS tiers; $1M needs ~350–400 labs — achievable but not trivial to sell.
Time to first revenue	10	8/10	Pilot-to-paid in weeks; owners feel the pain and pilots are cheap to run.
Defensibility	10	5/10	Per-lab master data + workflow lock-in compounds, but an incumbent LIS (Drlogy) could bolt this on. Speed and focus are the moat, not tech.
Total	100	72/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

A builder who can wrangle multimodal + STT pipelines and calibration, paired with someone who actually knows lab intake workflows (ex-lab-ops or LIS sales). Without the domain half, the master-mapping and escalation logic will be naive.

Key assumptions to validate (3–5)

1. Assumption: Order-entry mis-mapping (not physical sample errors) is a frequent, rupee-quantifiable pain for standalone labs. How to test: Sit in 10 labs for a day each; count intake corrections and re-draws traced to wrong-test-booked.
2. Assumption: Owners will pay ₹1,499–2,999/mo for an intake layer on top of their existing LIS. How to test: Pitch 30 labs the proof-video + price; measure pilot sign-ups and stated willingness.
3. Assumption: Confidence calibration is good enough that auto-pass saves time without introducing new silent errors. How to test: Shadow-run against 500 real orders; measure precision on auto-passed items vs. clerk baseline.
4. Assumption: Per-lab master onboarding can be done in <1 day per lab. How to test: Onboard 5 pilot labs and time it; find the bottleneck.

Risk flags

1. Platform/incumbent dependency: Drlogy or CrelioHealth could ship “AI order intake” as a free feature. Counter: go deep on multilingual/voice + aggregators before they notice, or partner as their intake layer.
2. Liability optics: A wrong auto-mapped test in a medical context is sensitive. Human-confirm-every-order and an audit log are non-negotiable design constraints, not add-ons.
3. Willingness-to-pay ceiling: Standalone labs are ruthless on ₹ line-items. If ROI isn’t screamingly obvious on the error-ledger, churn will be brutal.

14. Structured verdict

Score:                  72/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (multimodal + STT) + lab-ops/LIS-sales domain partner
Time to revenue:        6–10 weeks to first paid pilot
Capital to launch:      ₹6–12 lakh ($7–14K)
Top 3 assumptions to validate first:
  1. Mis-mapping is a weekly, rupee-quantifiable pain — shadow 10 labs, count re-draws from wrong-test-booked
  2. Owners pay ₹1,499–2,999/mo for an intake layer — proof-video pitch to 30 labs, measure pilot sign-ups
  3. Auto-pass precision beats clerk baseline — shadow-run 500 orders, measure silent-error rate
Kill criteria:
  - Abandon if <10% of 30 pitched labs start a pilot
  - Abandon if auto-pass introduces more silent errors than it prevents on the 500-order shadow test
  - Abandon if a top-2 LIS ships equivalent free intake AI before your v1 lands 20 paying labs

15. Next step — 1-week validation sprint

• Day 1–2: Build a throwaway demo: photograph 15 real (anonymized) prescriptions + record 10 Hinglish voice orders; run them through a Sarvam-STT + multimodal-LLM pipeline mapped to one sample lab master. Measure raw accuracy and where it’s unsure.
• Day 3–4: Take the demo to 8–10 nearby standalone labs and home-collection phlebs. Watch their intake for an hour each; count real mis-maps and re-draws. Show the demo mapping their messy script.
• Day 5: Decide go/no-go on a falsifiable bar: ≥4 of 10 labs verbally commit to a paid pilot AND at least one re-draw traced to wrong-test-booked observed per lab per week. Below that, the pain isn’t rupee-sharp enough — pass or re-cut toward the aggregator ops teams instead.